Definition
A subset of AI, which builds (trains) a predictive model from input data. Therefore, these AI systems are probabilistic.
What is machine learning?
Machine learning is a subset of artificial intelligence focused on algorithms that learn patterns from training data and use those patterns to make predictions or decisions about new data. Instead of relying on hard coded rules, a machine learning model improves its performance through experience.
The central idea is generalisation. A model is trained on a dataset designed to resemble real world problems. If training is successful, the model can apply what it has learned to new, unseen data. This use phase is called inference, where the trained model produces outputs such as predictions, classifications or decisions.
Machine learning now forms the backbone of most AI systems. It powers forecasting models, recommendation engines, computer vision, natural language processing, generative AI tools and many other applications.
How machine learning works
Machine learning works through mathematical models that map inputs to outputs. To do this, real world data must first be represented numerically. Each data point is typically encoded as a vector, where each dimension corresponds to a specific feature. Key mechanics include:
Features and data representation
Relevant characteristics of data, called features, are selected and transformed into numerical form. The process of selecting and preparing these features is known as feature engineering. Deep learning models often automate much of this process.
Model parameters and optimisation
Every machine learning model has adjustable parameters that control how inputs are weighted. During training, the model makes predictions, compares them to a target or feedback signal, measures error using a loss function, and updates parameters to reduce that error. The goal is to find parameter values that produce the most accurate results.
Training and generalisation
Training is not the final objective. The true goal is generalisation, meaning strong performance on real world data, not just on the training set. Evaluation methods such as train test splits and cross validation are used to check this.
Learning paradigms
Machine learning methods are commonly grouped into supervised learning, unsupervised learning, and reinforcement learning. Supervised learning uses ground truth targets, unsupervised learning finds patterns without labels, and reinforcement learning trains agents to maximise rewards through interaction with an environment.
Deep learning, which uses large neural networks with many layers, has become the dominant approach in many domains because it can model extremely complex patterns. Its rise has been closely tied to the availability of large datasets and powerful hardware such as GPUs.
A short history of machine learning
The term machine learning was coined in 1959 by Arthur Samuel, who described systems that could learn from experience, famously demonstrating a program that improved at playing checkers. Early conceptual foundations came from research into how neurons might work, including models proposed in the mid twentieth century that inspired artificial neural networks.
In the 1950s and 1960s, early learning machines were built to recognise patterns in signals such as speech or sonar. Research into pattern recognition and classification continued through the 1970s and 1980s, alongside growing interest in neural networks and learning algorithms. During this period, the field developed more formal definitions of learning systems and clearer mathematical frameworks for training models.
Over time, machine learning became more closely linked with statistics and optimisation, focusing on algorithms that improve performance with experience. The rise of large datasets and more powerful computing hardware in the late 2000s and 2010s enabled deep learning to achieve major breakthroughs. Neural networks with many layers began to outperform earlier approaches in areas such as computer vision and natural language processing, cementing machine learning as the dominant approach to building intelligent systems.
Machine learning vs artificial intelligence
Machine learning and artificial intelligence are closely related but not identical.
Artificial intelligence is a broad term for systems that can make decisions or predictions without ongoing human control. This can include simple rules-based systems, such as a thermostat that follows predefined if then rules.
Machine learning is a specific approach within AI where the system learns its behaviour from data rather than relying only on explicit rules.
All machine learning is AI, but not all AI uses machine learning.
As tasks become more complex, rules-based systems become difficult to scale because it is nearly impossible to manually encode every pattern and exception. Machine learning has therefore become the dominant method for building flexible, scalable AI systems.
Applications of machine learning
- computer vision tasks such as image classification and object detection
- natural language processing tasks such as chatbots, translation and text summarisation
- time series forecasting for markets, demand and anomaly detection
- recommendation systems in e commerce and media platforms
- generative models that create images, text or other content
- robotics and autonomous systems trained through reinforcement learning
These applications show how machine learning enables systems to analyse data, find patterns and act on those insights with limited human intervention.
Challenges and limitations
- Models can learn biases present in training data and reproduce unfair or harmful patterns.
- Many systems are difficult to interpret, leading to concerns about transparency and accountability.
- Models can also fail when exposed to data that differs from what they saw during training.
Other issues include privacy concerns, vulnerability to adversarial manipulation and the risk of overfitting, where a model performs well on training data but poorly in real world conditions. Ethical considerations such as fairness, regulation and responsible use are therefore central to modern machine learning practice.
Key takeaways
- Machine learning is a core subset of artificial intelligence focused on algorithms that learn from data rather than relying only on hard coded rules.
- Models are trained by adjusting parameters to reduce error and improve predictions, with generalisation to real world data as the main goal.
- The field has evolved from early pattern recognition systems to modern deep learning powered by large datasets and high-performance computing.
- Machine learning underpins major AI applications including computer vision, natural language processing, forecasting and generative AI.
- Challenges such as bias, interpretability, robustness and ethics make careful evaluation and governance essential.